Reducing Heap memory from 32 GB to less than 30.5 GB

(Vilas Reddy Podduturi) #1


I have my cluster setup with 32 GB per node with 5 nodes in ES 1.5 cluster. But now I read in the Elasticsearch documentation that heap memory is recommended to be 30.5 GB or less so that ES would use 32 bit addressing. So, I bounced one node at time with 30 GB heap memory. So does this change the addressing from 64 bit to 32 bit? Or it would work only with complete cluster restart?

Request your clarification.

(Mark Walkom) #2

No, it's a java pointer thing, but it is 32 bit.
Looks like the link to the oracle page that explained this it out of date, will see if we can get it updated.

(Vilas Reddy Podduturi) #3

Oh is it the case. So, now setting up the node with 30.5 GB or 32 GB does not matter?
As I mentioned we are not yet using Elasticsearch 2.0+. Currently using ES 1.5.

(Mark Walkom) #4

Yes it matters. Search for java compressed oops online.

(Vilas Reddy Podduturi) #5

So, thats great then. So it does not matter then if I did a full cluster restart or node by node for this to take affect right?

(Vilas Reddy Podduturi) #6

Also, is it enabled automatically or do I need to use the flag?
Because currently I restarted the nodes with 30 GB without any flag.

(Mark Walkom) #7

What flag? It just happens.
Just stay under the limit as the docs state.

(Vilas Reddy Podduturi) #8

I read online.. they talk about -XX:+UseCompressedOops flag.
But if it happens automatically, then its great.

(system) #9